[ehrvs]

Why: a key design principle of codex-local is that sc-lint enforces crate and feature independence between the fork-owned codex-rs/local/* crates and upstream, so frequent upstream merges stay trivial. An earlier commit had replaced sc-lint with a cargo metadata check. This PR restores sc-lint.

What:

• Adds codex-rs/sc-lint.toml and codex-rs/boundaries/ records for codex-local-{cli,models,resume}.
  • No upstream crate may depend on a local crate (allowed_dependents).
  • models and resume may only use their listed upstream crates.
  • cli may use codex-*, because it mirrors upstream cli's dependencies.
• SCB-FEATURE-001: no local crate may set features on an upstream crate.
• check_fork_surface.py drops its dependency-edge rule, which sc-lint now owns. It gains a coverage check: every codex-local-* crate must have a boundary record.
• CI installs sc-lint from feat(boundary): prefix patterns in dependency allow-lists and SCB-FEATURE-001 feature-independence rule randlee/sc-lint#603, which adds prefix patterns and SCB-FEATURE-001, at rev bc19f2b. There is a TODO to switch to randlee/sc-lint@v1 once that is released.

Verified locally:

• The lint passes with 0 findings.
• A temporary codex-core → codex-local-models edge fails with SCB-DEPENDENCY-002.
• A temporary features = [...] on an upstream dependency fails with SCB-FEATURE-001.
• Fork surface is ok at 131 lines over 12 files.

Depends on: randlee/sc-lint#603.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RPCzgwU247gMtjybR5v4iR

[ehrvs]

Superseded: merged into codex-local/main via the integration PR #21 (3908b3d).